View path planning for online 3D modeling using micro aerial vehicles

Abstract

This thesis addresses a view path planning problem to construct highly accurate 3D models using a Micro-Aerial-Vehicle. Most previous studies have focused on exploration approaches. They are a greedy strategy that iteratively determines the most informative view that exposes the largest unknown area from the current partial model. However, these approaches sometimes miss minor unreconstructed regions and produce unnecessarily long trajectories by revisiting already explored regions. Furthermore, they focus only on exploring the unknown area while disregarding the reconstruction quality

this may reduce the completeness and accuracy of reconstructed models. To address these issues, we provide a novel path planning method that utilizes an inspection strategy to model an unknown environment. Unlike conventional inspection problems, which assume that a prior structure is known, our method addresses online inspection according to partially known and consistently updated environments. The proposed method consistently computes an optimal inspection path in real-time, providing maximum coverage and minimum path length. Furthermore, we introduce a surface-based exploration method to analyze reconstructed surfaces in exploration planning. This method considers not only a volumetric model but also reconstructed surfaces using a real-time dense mapping algorithm. The method thoroughly analyzes the 3D graphical surfaces and efficiently explores the unknown region at the same time. The proposed methods are evaluated in comparison with other state-of-the-art exploration methods through simulated and real-world experiments. The results show that the proposed methods outperform the other methods and especially improve the completeness and accuracy of reconstructed 3D models.